How to summarize the context in Lagchain ReAct Agents

[Vedapani0402]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

tools = [Summary_tool, Time_series_Tool, ...]

llm = ChatOllama(model="mistral-small3.1:24b")

template = """You are a data analysis assistant......

You have access to the following tools:\
{tools}

Use the format below exactly. Do not skip or modify steps. Do not include any special symbols or extra formatting:

Question: the input question you must answer  
Thought: you should always think about what to do  
Action: the action to take, should be one of [{tool_names}]  
Action Input: the input to the action, if more than one value then just give the values alone separated with commas  
If there's an error, re-run the tool with corrected input  
Observation: the result of the action  
... (this Thought/Action/Action Input/Observation can repeat N times)  
Thought: I now know the final answer  
Final Answer: the final answer to the original input question

Begin!

Question: {input}  
Thought:{agent_scratchpad}

"""
prompt_template = PromptTemplate.from_template(template)

agent = create_react_agent(llm, tools, prompt_template)
agent_executor = AgentExecutor(
    agent=agent, 
    tools=tools, 
    handle_parsing_errors=True, 
    verbose=True,
)

Description

I am currently using Lagchain's ReAct Agent to dynamic analysis with tools defined. I am using Mistral Model from Ollama. The issue i am facing now is that after few tool calls, the agent is halucinating and not working properly.. Approximately after 10K Tokens the agent is failing to do the analysis it was supposed to and not even following ReAct Style mentioned in the prompt.

As the whole context of previous tool calls, tool data, prompt would be going as an input to the Agent, The context gradually increases. I want to know if there is a way to summarize the context for n tool calls and pass that summary instead of the whole context.

Langgraph has something called as SummarizationNode which does the same. Isn't there anything such in Langchain.

I tried using langraph ReAct agent with Open-Source Models but it did not work.

Can anyone help me with this.

Thank you

System Info

python version : 3.11
RTX A6000 48GB VRAM Linux System